BOREAL ENVIRONMENT RESEARCH 14: 499–514
ISSN 1239-6095 (print) ISSN 1797-2469 (online)
© 2009
Helsinki 31 August 2009
Comparison of net CO2 fluxes measured with open- and
closed-path infrared gas analyzers in an urban complex
environment
Leena Järvi1), Ivan Mammarella1), Werner Eugster2), Andreas Ibrom3),
Erkki Siivola1), Ebba Dellwik3), Petri Keronen1), George Burba4) and
Timo Vesala1)5)
1)
2)
3)
4)
5)
Department of Physics, P.O. Box 64, FI-00014, University of Helsinki, Finland
Institute of Plant Sciences, ETH Zürich, Universitätstrasse 2, CH-8092 Zürich, Switzerland
Bio Systems/Wind Energy Division, Risø National Laboratory, Danish Technical University (DTU),
Frederiksborgvej 399, DK-4000 Roskilde, Denmark
LI-COR Biosciences, 4421 Superior Street, P.O. Box 4425, Lincoln, NE 68504, USA
Department of Forest Ecology, P.O. Box 27, FI-00014 University of Helsinki, Finland
Received 28 Nov. 2008, accepted 7 Apr. 2009 (Editor in charge of this article: Veli-Matti Kerminen)
Järvi, L, Mammarella, I., Eugster, W., Ibrom, A., Siivola, E., Dellwik, E., Keronen, P., Burba, G. &
Vesala, T. 2009: Comparison of net CO2 fluxes measured with open- and closed-path infrared gas analyzers in an urban complex environment. Boreal Env. Res. 14: 499–514.
Simultaneous eddy covariance (EC) measurements of CO2 fluxes made with open-path
and closed-path analyzers were done in urban area of Helsinki, Finland, in July 2007–June
2008. Our purpose was to study the differences between the two analyzers, the necessary
correction procedures and their suitability to accurately measure CO2 exchange in such
non-ideal landscape. In addition, this study examined the effect of open-path sensor heating on measured fluxes in urban terrain, and these results were compared with similar
measurements made above a temperate beech forest in Denmark. The correlation between
the two fluxes was good (R2 = 0.93) at the urban site, but during the measurement period
the open-path net surface exchange (NSE) was 17% smaller than the closed-path NSE,
indicating apparent additional uptake of CO2 by open-path measurements. At both sites,
sensor heating corrections evidently improved the performance of the open-path analyzer
by reducing discrepancies in NSE at the urban site to 2% and decreasing the difference in
NSE from 67% to 7% at the forest site. Overall, the site-specific approach gave the best
results at both sites and, if possible, it should be preferred in the sensor heating correction.
Introduction
Due to the growing concern about changing
climate, there has been an increasing interest to
measure the exchange of carbon dioxide (CO2)
between terrestrial ecosystems and the atmosphere. For this purpose, direct measurements of
CO2 exchange with the eddy covariance (EC)
technique have been widely used (e.g. Aubinet
et al. 2000, Baldocchi 2003). Traditionally, EC
measurements have been made above vegetated
surfaces, but were recently extended to urban
areas, which play an important role as a source
of CO2 (e.g. Grimmond et al. 2002, Nemitz et
500
al. 2002, Moriwaki and Kanda 2004, Vogt et al.
2006, Vesala et al. 2008). The urban environment brings its own challenges to EC measurements due to the complexity of the surface,
which may not always provide homogeneous
and stationary conditions.
Most commonly, CO2 fluxes are being measured with either open- or closed-path CO2/H2O
infrared gas analyzers in combination with fast
response sonic anemometers. The measurement
principles of these two types of analyzers are
rather different, and they may not be equally
suitable for measurements in urban environments. An open-path analyzer measures partial
CO2 densities (μmol m–3) in situ in ambient
air, which can be directly used for computing
fluxes (in unit μmol m–2 s–1) via the covariance
approach. This apparent advantage is, however,
hampered by the fact that any variation in the
density of air (caused e.g. temperature or humidity fluctuations) in the sampled air volume that
is not perfectly uncorrelated with the variations
in partial CO2 densities leads to apparent fluxes
that need a special correction (see below). In
contrast, closed-path analyzer uses a controlled
sample air volume with highly reduced temperature fluctuations at a stabilized temperature
and known pressure. Depending on whether the
air was dried or water vapour concentrations
were measured simultaneously, such instruments
could measure dry air mole fractions (or mixing
ratios) in ppm. Measuring dry mole fraction
requires, however, the density of air in eddy
covariance flux computations to yield flux densities in μmol m–2 s–1. This is usually not measured
with a very high accuracy.
Fluxes measured with both systems need to
be corrected for inadequate system response,
which has an especially large effect on closedpath measurements. Often these frequency
response corrections are based on a model cospectra obtained above a flat terrain (e.g. Kaimal
et al. 1972, Eugster and Senn 1995). The usage
of these model co-spectra may cause an error in
the frequency response corrections especially in
complex terrains, where co-spectra may have a
different shape. Spectral studies made in urban
areas have so far reported similar co-spectral
shapes as those by Kaimal et al. (1972) at sev-
Järvi et al.
•
BOREAL ENV. RES. Vol. 14
eral specific locations (e.g. Roth and Oke 1993,
Roth 2000, Vesala et al. 2008). The CO2 fluxes
measured with open-path analyzers need also to
be corrected for the density fluctuations caused
by correlated heat and water vapour fluctuations (WPL correction, Webb et al. 1980). The
unresolved energy budget closure observed in
many flux sites is likely to affect the accuracy
of the WPL correction (e.g. Liu et al. 2006). The
energy imbalance can originate from advection
of both heat and moisture which are assumed to
be higher in urban areas due to the heterogeneity of the surface. There is also a concern about
pressure variance term in the WPL correction
which is usually assumed to be negligible, but
which might have significant bias at the sites
having frequently high wind speeds and strong
turbulence (Kramm et al. 1995, Massman and
Lee 2002). Lately, there has also been a discussion about the heat flux produced by the openpath sensor itself, which should be taken into
account when correcting CO2 fluxes (Burba et
al. 2006, Grelle and Burba 2007, Burba et al.
2008). Previous studies have reported the surface
heating causing an apparent CO2 flux outside the
growing season (Grelle and Burba 2007, Burba
et al. 2008, Ono et al. 2008).
The main purpose of this study is to compare
CO2 fluxes measured with open-path and closedpath CO2/H2O infrared gas analyzers (LI-7500
and LI-7000, respectively) in a complex urban
environment in Helsinki, Finland. The measurement location creates rather different ground for
the CO2 exchange, with high emissions through
the year covering all the four seasons. So far,
comparisons between these two analyzers have
been made under controlled conditions or above
vegetated surfaces (Leuning and King 1992,
Suyker and Verma 1993, Grelle and Burba 2007,
Burba et al. 2008, Ono et al. 2008). Comparisons
are lacking in urban areas, in which also errors
concerning the technique itself are expected to
be larger. The secondary purpose is to examine
the sensor heating correction of the open-path
analyzer, and correction methods proposed by
Burba et al. (2006, 2008) in an urban environment alongside the measurements made above
forest located in Lille Bøgeskov, Denmark, are
tested.
BOREAL ENV. RES. Vol. 14 •
Net CO2 flux comparison in an urban environment
Materials and methods
Measurement sites and instrumentation
The measurements were carried out at the
SMEAR III (Station for Measuring Ecosystem–
Atmosphere Relationships) station in Helsinki,
Finland, located about 5 km to the north-east
from the centre of Helsinki. EC measurements
were made on the top of a 31-meter-high lattice
tower (60°12.17´N, 24°57.671´E in WGS84) situated on a rocky hill 26 meters above sea level.
A maximum slope of 12° is present in the direction 120°–180° (Vesala et al. 2008). Besides
the tower, meteorological measurements were
placed on the roof of the University of Helsinki
buildings located next to the measurement tower.
The surrounding area is heterogeneous consisting of buildings, paved areas, patchy forest and
cultivated area, and it can be divided into three
land use sectors (urban 320°–40°, road 40°–180°
and vegetation 180°–320°), each representing
typical land use in the area. A more detailed
description about the station can be found from
Järvi et al. (2009) and Vesala et al. (2008).
The EC setup at the SMEAR III station to
measure CO2 fluxes consisted of a Metek ultrasonic anemometer (USA-1, Metek GmbH, Germany) to measure all three wind components
and sonic temperature, as well as open- and
closed-path infrared gas analyzers (LI-7500
and LI-7000, respectively, LI-COR, Lincoln,
Nebraska, USA) to give CO2 and water vapour
densities and mixing ratios, respectively. The EC
setup was situated on a boom, with a distance of
1.3 meters to the south-west of the tower. The
open-path analyzer was mounted 0.2 m away
from the anemometer and it was tilted 30° in
the north-west direction to allow rainwater to
drip off. In case of the closed-path analyzer, air
sample was taken 13 cm below the anemometer and it was drawn to the analyzer through a
40-meter-long steel tube with an inner diameter
of 8 mm. The flow rate in the tube was 17 l min–1
to ensure turbulent flow, and the sampling line
was heated with a power of 4 W m–1 to avoid
condensation of water vapour. The open-path gas
analyzer was connected to the analogue input
channels of the anemometer, whereas closed-path
501
analyzer data were recorded separately using the
RS-232 output. The frequency of the EC measurements was 10 Hz and the raw data were stored
for later post-processing.
Besides the EC measurements, the air temperature was measured with a Pt-100 resistance (“home-made”, ventilated and shielded
from rain and solar radiation) thermometer, and
incoming and outgoing shortwave and longwave radiation with a net radiometer (CNR1,
Kipp & Zonen, Delft, Netherlands) at the height
of 31 meters. The measurement interval for
these was 1 minute. Air pressure was measured
with a barometer (Vaisala DPA500, Vaisala Oyj,
Vantaa, Finland) on the top of the University of
Helsinki building at the height of 50 meters with
a measurement interval of 4 minutes.
Similar open-path and closed-path CO2 flux
measurements were also conducted over natural
ecosystem for comparison with the urban environment. The measurements were made above
a beech forest in Lille Bøgeskov (55°29´N,
11°38´E, 40 m a.s.l.) near Sorø in Zealand,
Denmark (Pilegaard et al. 2003). The EC measurements were made at the height of 43 meters
and the setup consisted of a Solent ultrasonic
anemometer (R2, Gill Instruments, Lymington,
UK), and LI-7500 and LI-7000 infrared gas analyzers. The open-path analyzer was tilted 10–15
degrees. In the case of the closed-path analyzer, air was sampled 0.6 m distance below the
anemometer and was drawn through a 50 meters
long sampling tube to the analyzer with inner
diameter of 8 mm and flow rate of 20 l min–1.
The sampling frequency was 10 Hz.
Basic equations and data processing
The 30-min turbulent flux of scalar s was calculated as a co-variance between the vertical
wind speed and s according to e.g. Aubinet et
al. (2000). Before the flux calculation, the linear
trend was removed from the time series and twodimensional coordinate rotation was applied.
Flux values were calculated with a maximum
covariance method (see McMillen 1988, Moncrieff et al. 1997) which also gave a time lag of
7.3 s for closed-path CO2 caused by the tube.
502
Järvi et al.
Below we go through the other necessary corrections in a more detail.
Co-spectral corrections
The atmospheric fluxes measured with both openpath and closed-path sensors have flux losses at
both high and low-frequency ends. At low frequencies, the losses are caused by the averaging
time and linear de-trending used in the flux calculations, while at the high frequency end losses are
due to the imperfect frequency response of the
EC system (e.g. Moore 1986, Eugster and Senn
1995, Aubinet et al. 2000, Rannik et al. 2004).
The underestimation of the measured flux can be
determined with the following equation:
,
(1)
where f is the natural frequency (in Hz), Fm and
F are the measured and un-attenuated scalar
fluxes, TFL(f ) and TFH(f ) are the transfer functions accounting for the low- and high-frequency
losses, and Cws(f ) is the co-spectral density
describing the frequency behaviour of the turbulent flux (Horst 1997, Moore 1986, Massman
2000).
At the urban site, the co-spectral corrections
were calculated by integrating numerically Eq. 1.
By assuming a scalar similarity, co-spectral
models for sensible heat by Kaimal et al. (1972)
were used as Cws(f ), and the normalized frequency nm in which the co-spectrum of sensible
heat attains its maximum value was determined
separately for each land use sector (e.g. Kaimal
et al. 1972, Horst 1997). In an unstable case, nm
was independent from wind direction and stability with an average value of 0.10 ± 0.04 (SE).
In stable cases, the stability dependence of nm
differed among the land-use sectors with the best
approximations determined by nonlinear fitting in
a least-square sense (Rannik et al. 2004):
,
(2)
•
BOREAL ENV. RES. Vol. 14
In Eq. 2, ζ = (z – d)/L is the stability parameter, where z is the measurement height, L is the
Monin-Obukhov length and d is the displacement height (13, 8 and 6 m for urban, road and
vegetation sectors, respectively) approximated to
be 2/3 of the mean canopy height.
For the open-path EC system, TFH(f ) was
calculated as the product of the appropriate theoretical transfer functions associated with the
high-frequency attenuation according to Moncrieff et al. (1997). The co-spectral transfer function TFH(f ) for the closed-path EC system was
experimentally calculated according to
,
(3)
where τc is the first order response time (s)
defined from the measured co-spectra of sensible
heat and CO2 fluxes. For the low frequency loss,
the functional form of the co-spectral transfer
function TFL(f ) associated with linear detrending
was used for both systems (Rannik and Vesala
1999).
The detailed spectral corrections used at the
forest site are presented in Ibrom et al. (2007).
WPL correction
The fluxes measured with the open-path analyzer
were corrected for temperature and water vapour
fluctuations according to Webb et al. (1980):
, (4)
where Fc and F0 are the WPL-corrected and
co-spectrally corrected open-path CO2 fluxes,
respectively, (in kg m–2 s–1), μ is the ratio of
molar masses of air and water vapour (μ =
1.6077), rc, rv and rdry are the densities of CO2,
water vapour and dry air (in kg m–3), respectively, Ta is the air temperature,
is the
water vapour flux (in kg m–2 s–1) and
is the
sensible heat flux (in K m s–1) corrected for cross
wind and humidity effects as proposed by Schotanus et al. (1983). In Eq. 4, rc and rv obtained
from closed-path analyzer were used, since density measurements made with open-path analyzers may be more inaccurate due to e.g. lens
contamination (Serrano-Ortiz et al. 2008).
BOREAL ENV. RES. Vol. 14 •
Net CO2 flux comparison in an urban environment
Effect of open-path sensor heating on WPL
correction
The effect of the density fluctuations generated
by the open-path analyzer itself should be taken
into account in the traditional WPL correction.
Burba et al. (2006, 2008) proposed three slightly
different correction procedures for this sensor
heating and all these procedures are tested in
practice here.
In Burba et al. (2006), an additional term is
added to FOP calculated with Eq. 4, so that the
final corrected flux, FFit, can be presented as
, (5)
where δ is that fraction of heat produced by the
analyzer which stays in the thermal boundary
layer of the analyzer and is relevant for the correction, Ts is the instrument surface temperature
(°C) and
is the aerodynamic transfer
resistance (s m–1) (Jones 1992). Here the unit for
density of CO2 is μmol m–3. By assuming a linear
relationship (Ts = a0 + a1Ta) between Ts and Ta,
and that the closed-path CO2 flux is the true flux
FFit, Eq. 5 can be solved as a least-square regression to obtain parameters δ, a0 and a1. In the
fitting, initial guesses for all the unknown variables were used. For δ, a value of 0.05 suggested
by Rogiers et al. (2008) was used, while for a0
and a1, the bottom window values separately for
day and night from the body heating experiment
from Burba et al. (2008) were used. We used
bootstrapping to obtain more generalized correction parameters separately for daytime and
nighttime at both urban and forest sites. In bootstrapping, the data were divided into 100 subsamples, each including arbitrary 5/6 of the data.
The parameters δ, a0 and a1 were solved for each
subset and the final ones to correct the open-path
data were calculated as arithmetic means from
the subset parameters (Table 1). The sensitivity
of the parameters was tested by choosing only
2/6 of the data for each 100 subsample. This
affected less than 3% to the parameters δ, a0
and a1 at both sites. FFit was then calculated with
the fitted parameters. Henceforth this method is
referred to as the fitting method (Fit).
The fitted value of Ts describes the combined
surface temperature of the different parts of the
503
analyzer, which are influenced by the instrument
electronic and prevailing meteorology (especially solar radiation). Heating caused by the
electronics increases with decreasing Ta, while
that caused by the sun increases with increasing
Ta. In a daytime forest, Ts decreases with increasing Ta similarly as was observed by Burba et al.
(2008), and thus the heating by the electronics
dominated Ts (Table 1). However, Ts increases
with increasing Ta at the urban site probably
due to heating by the sun. The larger effect from
solar heating is likely related to the position of
the analyzer, since tilting affects how much different parts of the analyzer are subjected to solar
radiation. Also the shorter measurement period
in the forest with different radiation conditions
possibly affects the observed difference between
the forest and urban sites.
In Burba et al. (2008), the sensor heating correction is done by adding the heat flux generated
by the sensor to the heat flux measured with the
sonic anemometer, and the sensor heating corrected flux is calculated with Eq. 4. The sensible
heat flux caused by the analyzer itself can be
calculated via modified resistance approach following Nobel (1983) when the surface temperatures of the open-path analyzer are known. If the
surface temperatures are not measured, Burba et
al. (2008) proposed two methods to calculate the
surface temperatures of different parts (bottom
window, upper window and spar) of the LI-7500.
In the first method, the difference between Ts and
Ta is calculated using the following equation:
, (6)
where x stands for bottom window, upper
window and spar, Rg is the global radiation
(W m–2), RIR is the incoming long-wave radiation (W m–2), U is the wind speed (m s–1)
Table 1. Parameters (± SD) δ, a0 and a1 obtained with
the fitting method.
δ
Daytime
Nighttime
a0
a1
a0
a1
Urban
Forest
0.060 ± 0.011
1.77 ± 0.07
1.14 ± 0.01
–0.38 ± 0.040
1.13 ± 0.01
0.085 ± 0.003
3.17 ± 0.17
0.93 ± 0.01
1.52 ± 0.14
1.05 ± 0.02
504
Järvi et al.
fcov CO2 flux(f)/w´s´
100
a
10–2
sensible heat
CO2 fluxopen
CO2 fluxclosed
theoretical (Kaimal et al. 1972)
Co-spectral transfer function
10–4
10–3
10–2
10–1
n = f(z − d)U–1
100
101
b
1
0.5
CO2
0
τc = 0.55s
10–2
10–1
Frequency (Hz)
100
Fig. 1. (a) Normalized logarithmic co-spectra of the
sensible heat flux and CO2 fluxes obtained from openpath and closed-path systems as a function of normalized frequency n. (b) effective transfer function of
closed path CO2 versus natural frequency f. τc is the
first order response time determined from effective
transfer function. Co-spectra are calculated as an average from 270 half-hour unstable co-spectra between
March 2008 and June 2008.
and b0,x … b4,x are parameters obtained with
the multiple linear regression analysis made for
hourly surface temperature measurements (see
table 2 in Burba et al. 2008). In the second
method, the surface temperatures of the bottom
window, upper window and spar are calculated
linearly from Ta using parameters obtained from
hourly measurements in the body heating experiment (see table 3 in Burba et al. 2008). Below,
these two methods utilizing a modified resistance approach to correct for the sensor heating
are called the multiple linear regression method
(MLR) and the linear method (Lin), and the
open-path CO2 fluxes obtained with these methods are denoted by FMLR and FLin, respectively.
Analyzed periods and data removal
We analyzed the data over one year from the
urban site during July 2007–June 2008. During
this period, data gaps due to maintenance and
•
BOREAL ENV. RES. Vol. 14
power outages affected 12% of all the flux measurements. In the quality control step, clear spikes
identified by visual inspection were removed,
and half-hour fluxes with a friction velocity u*
< 0.1 m s–1 were omitted. To ensure stationary
conditions, a stationary test according to Foken
and Wichura (1996) was applied. In the test,
30-minute intervals used in the flux calculation
were divided into six five-minute sub-intervals,
and if the difference between the 30-minute covariance and the mean covariance of the sub-intervals was greater than 30%, the flux value was
removed. Overall, the quality analysis removed
31% and 44% of closed-path and open-path CO2
flux data, respectively. The higher rejection of the
open-path data was due to the larger amount of
the data spikes caused by rain or fog. Comparisons between the CO2 fluxes obtained with openpath and closed-path analyzers were made only
when the data from both analyzers were available
after the quality control procedure. This reduced
the data to 7060 half-hour data records, covering
40% of the whole year.
The data from the beech forest in Denmark
were collected between October and December
2006. After quality control procedures similar
to those used for the urban site, the remaining
forest data covered 41% from the three month
period.
Results and discussion
Results below concern the measurements made
at the urban site in Helsinki, except in the last
section where the effect of different surface heating corrections on FOP are presented for urban
and forest sites.
Frequency response corrections
The average normalized co-spectrum of sensible heat (Fig. 1a) follows the Kaimal et al.
(1972) co-spectra, suggesting that the standard
co-spectral model was applicable at our urban
measurement site. The open-path CO2 flux had
only a slight effect due to the inadequate frequency response, whereas the effect of highfrequency attenuation was distinct in the normal-
Net CO2 flux comparison in an urban environment
ized co-spectrum of the closed-path CO2 flux.
The first-order response time τc to be used in
Eq. 3 was determined experimentally from the
ratio of co-spectral density of CO2 and sensible
heat (Rannik et al. 2004), which was assumed
to be free from attenuation. From the effective
co-spectral transfer function of the closed-path
system, a value of τc = 0.55 s was found for the
response time (Fig. 1b).
On average, the co-spectral flux losses were
11% ± 3% (SD) and 3% ± 2% (SD) for closedand open-path CO2 fluxes, respectively. The flux
loss of both measurement systems depended on
stability and wind speed similarly as reported by
Moore (1986) and Massman (2000).
Correlation between the open- and
closed-path CO2 fluxes
The CO2 flux obtained with the closed-path
analyzer (FCP) was correlated with FOP for the
period July 2007–June 2008 (Fig. 2). A correlation analysis yielded R2 = 0.93, showing
that the two setups were measuring the same
signal. The slope using a linear least square fit
was slightly lower than one. From the offset
(0.61 μmol m–2 s–1) it can be seen that the openpath analyzer gave smaller CO2 fluxes than the
closed-path analyzer, which corresponds to an
apparent additional uptake of CO2. This was
mainly due to the daytime situations (FCP =
0.97FOP – 1.39). The nocturnal fit (FCP = 0.90FOP
+ 0.40) gave a slightly positive offset, showing
that the open-path analyzer was measuring larger
fluxes than the closed-path analyzer. In general,
the daytime correlation was better than the nocturnal one. With the correlation between the two
fluxes being so high, even a small difference of
the regression slope from 1.0 and the offset from
0.0 will have a large effect on the annual carbon
budgets.
The correlation found in this study is better
than the observed correlations of R2 = 0.79 and
R2 = 0.70 between two closed-path systems with
separate anemometers above arctic tundra and
forest, respectively (Eugster et al. 1997, Rannik
et al. 2006). Burba et al. (2008) found a weaker
correlation (R2 = 0.86) between the two sensors with the same anemometer above grassland
505
80
FOP (μmol m–2 s–1)
BOREAL ENV. RES. Vol. 14 •
60
Data
1:1
40
20
0
–20
–20
y = (0.96 ± 0.01)x + (−0.61 ± 0.05)
R2 = 0.93
0
20
40
60
80
FCP (μmol m–2 s–1)
Fig. 2. Correlation (p < 0.05) between the urban CO2
fluxes measured with closed-path (FCP) and open-path
(FOP) analyzers between July 2007 and June 2008. The
equation gives a least-square linear regression and
95% CIs.
during winter with the slope of 0.82 and the
offset of –0.45 μmol m–2 s–1. A better correlation
(R2 = 0.98) was found by Ibrom et al. (2007)
above a beech forest in summer (slope 1.00,
offset 0.44).
Diurnal cycle of FCP and FOP
The diurnal behaviours of FOP and FCP were plotted separately for different temperature ranges
(Fig. 3) due to the dependence of the difference
ΔFC = FCP – FOP on Ta (Fig. 4). The diurnal pattern was calculated as medians from half-hour
flux points.
The diurnal patterns of FCP and FOP followed
closely each other. The surroundings acted as a
source of CO2, except in the highest temperature
range of 12–28 °C when downward (negative)
fluxes were observed during daytime. The highest downward fluxes (–8 μmol m–2 s–1) were due
to the direction of prevailing winds (SW) which
was also the direction of the highest vegetation cover. As suggested already by the correlation analysis, the closed-path analyzer measured
higher upward and lower downward (depending
on Ta) fluxes than the open-path analyzer during
daytime. Below 5 °C, the difference between the
two fluxes was around 0.8 μmol m–2 s–1 (10% difference), increasing to 3.5 μmol m–2 s–1 (25% difference) in the temperature range of 12–28 °C.
This indicates that the underestimation of the
WPL correction increases with the increasing
506
Järvi et al.
20 a
•
BOREAL ENV. RES. Vol. 14
b
10
Fc (μmol m–2 s–1)
0
–10
20 c
d
10
0
–10
0
6
12
Time (EET)
18
24 0
6
FCP
0.5
12
Time (EET)
18
24
FOP
Fig. 3. Median diurnal variation of urban CO2 fluxes
obtained with closed-path
(FCP) and open-path (FOP)
analyzers for temperature ranges (a) –12 to
0 °C (b) 0 to 5 °C, (c) 5
to 12 °C, and (d) 12 to
28 °C. Values were calculated from the half-hour
flux points and grey lines
show the respective ±1
quartiles.
a
0
–0.5
–1
∆Fc |FCP|–1
0.5
b
0
–0.5
–1
0.5
c
0
–0.5
–1
–10
–5
0
5
10
Ta (°C)
temperature. The most likely reason for this
was the heating of the open-path sensor surface
caused by stronger solar radiation at higher temperatures. Previous studies have also found that
the open-path analyzer is measuring 10%–20%
lower upward/higher downward CO2 fluxes than
the closed-path analyzer (e.g. Leuning and King
1992, Liu et al. 2006, Ono et al. 2008). The
15
20
25
Fig. 4. Temperature
dependence of the difference (ΔFC = FCP – FOP)
between CO2 fluxes measured with open-path (FOP)
and closed-path (F CP)
analyzers normalized with
FCP in the urban area (a)
for the whole measurement period between July
2007 and June 2008, and
separately for (b) daytime
and (c) nighttime fluxes.
Values are medians calculated from half-hour data.
reason for this has been reported to be underestimation of the WPL correction due to an unbalanced energy closure (Ham and Heilman 2003,
Liu et al. 2006). Some studies have found the
heating of the open-path sensor to be the reason
for the apparent ecosystem uptake outside the
growing season (Grelle and Burba 2007, Burba
et al. 2008, Ono et al. 2008).
Net CO2 flux comparison in an urban environment
The closed-path analyzer measured most of
the time higher fluxes than the open-path analyzer when upward fluxes were observed. Nevertheless, during cold nights (below 5 °C), FCP
reached 0.7 μmol m–2 s–1 lower values than FOP,
corresponding to a difference of 10%–30%. This
is propably related to different meteorological
conditions when outgoing long-wave radiation is
able to cool the surface of the analyzer below the
ambient temperature. However, also cold nocturnal fluxes include several possible error sources
and thus should be considered with caution. First
of all, the open-path analyzer may be facing
problems at cold temperatures because no external heater was used. On the other hand, the anemometer heater turns on at 5 °C, which might
cause an overestimation of the WPL correction.
Problems may also be related to the nocturnal
functionality of the closed-path system, such as
inadequate frequency response corrections due
to the deviations from model co-spectrum. More
generally, Ono et al. (2008) stated that the conventional WPL correction might be inappropriate in situations when this corrections is of same
order or greater than the raw FOP.
Systematic differences between FCP and
FOP due to meteorological variables
Beside temperature, the difference between the
two fluxes had systematic dependence on the
wind speed (U), wind direction (WD) and the
variable ζ (Fig. 5). In all temperature ranges, the
normalized ΔFC increased from near zero to 0.8
with an increasing wind speed, especially at U >
8 m s–1 (Fig. 5a). The data points with U > 10
m s–1 should be interpreted with caution due to
their low number (fewer than 12) in the bin U =
10–12 m s–1. At strong wind speeds, the pressure term in the WPL correction might become
important (Massman and Lee 2002), and ignoring it should lead to an underestimation of FOP as
was found in our case. This, however, disagrees
with Ono et al. (2008) who found that ignoring
the pressure variance term could not explain the
difference between CO2 fluxes measured with
open- and closed-path analyzers above a rice
paddy. An alternative interpretation for this U
507
1
0.5
0
–0.5
0
2
4
6
U (m s–1)
8
10
12
1
∆Fc |FCP|–1
BOREAL ENV. RES. Vol. 14 •
0.5
0
–0.5
0
45
90
135
180
225
WD (°)
270
315
360
1
0.5
0
–0.5
–1
–0.5
Raw data
–12 < Ta < 0
0
ζ
0.5
1
0 < Ta < 5
5 < Ta < 12
12 < Ta < 28
Fig. 5. The median dependence of ΔFC normalized
with FCP on (a) wind speed U (m s–1), (b) wind direction WD (°) and (c) stability ζ in different temperature
ranges at urban site from July 2007 to June 2008. Grey
points show the half-hour ratios. The limits of different
land-use sectors (urban, road and vegetation) are also
plotted.
dependence could be errors in sonic temperature
measurements at high wind speeds, causing an
underestimation of the sensible heat flux (Grelle
and Lindroth 1996), which affects the WPL correction and further FOP.
The effect of complex measurement surroundings can be seen in a highly variable WD
dependence of the normalized difference (Fig.
5b). The most distinguishable feature is the peak
of normalized ΔFC that reached the value of 0.4
in the direction 190°–250° that corresponds to
the vegetation sector. At the same time, the normalized ΔFC decreased to –0.25 with a decreasing temperature. This suggests that the peak was
related to the vegetation cover, which caused an
underestimation of the WPL term. The measure-
508
Järvi et al.
2500
•
BOREAL ENV. RES. Vol. 14
FCP
Cumulative Fc (g C m–2)
FOP
2000
1500
1000
500
0
Aug 2007
Oct 2007
Dec 2007
Feb 2008
ment tower located in the direction 10°–50° did
not have any distinct effect on the normalized
ΔFC. Tilting of the analyzer could have the effect
that flows coming from certain directions relative to the analyzer would remove the heat more
efficiently from the optical path of the analyzer
than flows coming from some other directions.
This kind of phenomenon was, however, not
observed. Normalized ΔFC was larger under
unstable conditions than under stable conditions,
except in the temperature range of –12 to 0 °C
(Fig. 5c) where the small positive daytime and
negative nighttime normalized ΔFC cancelled
each other out. Larger differences observed
under unstable conditions support the observed
diurnal behaviour.
Annual net surface exchange (NSE)
Despite the close relation between the CO2
fluxes measured with the closed-path and openpath analyzers during July 2007–June 2008 (Fig.
2), the analyzers gave a substantially different
net surface exchange (NSE), with values of 2630
and 2180 g C m–2 measured with closed-path and
open-path analyzers, respectively (Fig. 6). Thus,
measurements made with the open-path analyzer
would underestimate the total carbon emission
by 450 g C m–2, corresponding to a difference of
17% between the analyzers. Burba et al. (2008)
reported a yearly difference of 14% to 16% in
the NSE measured above maize and soybean,
which is comparable to our result, even though
the measurement environments were rather different.
Apr 2008
Jun 2008
Fig. 6. Cumulative urban
CO2 fluxes (g C m–2)
obtained with closed-path
(FCP) and open-path (FOP)
analyzers from July 2007
to June 2008. No gap filling was made.
The effect of open-path sensor heating
on CO2 fluxes
The effect of open-path sensor heating on the
measured CO2 fluxes was studied at both urban
and forest sites. Comparisons between the different CO2 fluxes were made when all five fluxes
(FCP, FOP, FFit, FMLR and FLin) were available,
giving a total of 5641 and 1806 half-hour data
points at urban and forest sites, respectively.
Diurnal behaviour of the corrected CO2
fluxes
Below 5 °C, the linear method overcorrected the
open-path flux over the whole diurnal course,
while the fitting and MLR methods removed
most of the difference in daytime (Fig. 7). At
night, however, also the MLR method failed
to correct FOP by overcorrecting it. The fitting
method did not notably affect nocturnal ΔFC
in the temperature range 0–5 °C (Fig. 7b), but
lowered nocturnal FOP in the coldest temperature
range of –12 to 0 °C, and ΔFC approached 0 (Fig.
7a). The fitting method is the only method that
takes into account the magnitude of nocturnal
cooling well enough (see also Table 1), while
with the MLR and linear methods this high surface cooling is not possible. In temperature range
5–12 °C, all correction methods improved the
open-path flux, but the best results were obtained
with the linear method especially at night (Fig.
7c). The fitting and linear methods corrected the
nocturnal open-path fluxes well in the temperature range 12–28 °C (Fig. 7d), whereas the fitting
BOREAL ENV. RES. Vol. 14 •
Net CO2 flux comparison in an urban environment
4
a
b
c
d
509
2
4
2
0
–2
0
6
12
Time (EET)
method was best in daytime between 08:00–
17:00. In early morning and evening, the fitting
and linear methods overestimated the open-path
flux because the linear estimation of Ts from Ta
yielded too high values of Ts. During the transitions from night to day and day to night, the
surface of the analyzer likely heats and cools,
respectively, slower than the ambient air due to
the solar radiation. This was observed only in the
highest temperature range when there is strong
enough solar radiation in Helsinki.
The diurnal behaviour of closed-path and
open-path CO2 fluxes corrected in the traditional way were plotted for temperature ranges
0–12 °C and 12–18 °C above forest (see Fig. 8).
There were only few data points below 5 °C, so
the temperature range 0–5 °C was not separated.
Similarly to urban site, FCP was larger/smaller
than FOP when the flux was upward/downward.
The apparent net ecosystem uptake was visible
in the colder temperature range when FCP was
zero but FOP was –2 μmol m–2 s–1 in daytime
similarly to previous studies made above vegetation outside the growing season (Grelle and
Burba 2007, Burba et al. 2008, Ono et al. 2008).
All the three correction methods improved the
CO2 flux measured with the open-path analyzer,
18
24 0
6
12
Time (EET)
FCP – FOP
FCP – FMLR
FCP – FLin
FCP – FFit
18
24
a
5
0
–5
Fc (μmol m–2 s–1)
Fig. 7. Median diurnal
behaviour of the differences (ΔF C) between
CO2 fluxes (μmol m–2 s–1)
obtained with closed-path
and open-path analyzers
at the urban site for different temperature ranges.
FCP is the closed-path
flux, FOP is the traditionally corrected open-path
flux, and FMLR, FLin and FFit
are the open-path fluxes
corrected with the MLR,
linear and fitting method,
respectively (see text for
details).
∆Fc (μmol m–2 s–1)
0
–10
b
5
0
–5
–10
–15
0
6
12
Time (EET)
FCP
18
24
FOP
Fig. 8. Median diurnal variation of CO2 fluxes measured
with closed-path (FCP) and open-path analyzers (FOP)
above beech forest in Denmark for temperature ranges
(a) 0–12 °C, and (b) 12–28 °C in October–December
2006.
but the best results were obtained with the fitting method especially in the temperature range
0–12 °C (Fig. 9). ΔFC did not exhibit a distinct
diurnal pattern above forest (Fig. 9) contrary
to the urban site where the diurnal pattern was
510
Järvi et al.
4
2000
a
•
BOREAL ENV. RES. Vol. 14
a
1500
2
Cumulative Fc (g C m–2)
1000
Fc (μmol m–2 s–1)
0
–2
4
b
2
500
0
0
200
1000
2000
3000
4000
5000
b
150
100
50
0
0
–50
–2
0
6
12
18
Time (EET)
FCP – FOP
FCP – FLIN
FCP – FMLR
FCP – FFIT
0
500
24
Fig. 9. Median diurnal behaviour of the differences
(ΔFC) between CO2 fluxes (μmol m–2 s–1) obtained with
the closed-path and open-path analyzers above the
forest for different temperature ranges [(a) 0–12 °C,
and (b) 12–28 °C] in October–December 2006. FCP is
the closed-path flux, FOP is the traditionally corrected
open-path flux, and FMLR, FLin and FFit are the openpath fluxes corrected with the MLR, linear and fitting
method, respectively (see text for details).
pronounced even after the surface heating corrections. This is likely due to the shorter measurement period at the forest when the radiation
conditions were rather different than those at the
urban site over the whole summer.
1000
Running index
FCP
FOP
FLIN
FFIT
1500
FMLR
Fig. 10. Cumulative CO2 fluxes (g C m–2) measured
with closed-path and open-path analyzers (a) at urban
site from July 2007 to June 2008, and (b) above forest
in October–December 2006. FCP is the closed-path flux,
FOP is the traditionally corrected open-path flux, and
FMLR, FLin and FFit are the open-path fluxes corrected
with the MLR, linear method and fitting method, respectively (see text for details). No gap filling was made.
Cumulative CO2 fluxes
We calculated cumulative CO2 fluxes with different methods for both urban and forest sites (no
gap filling was made) (Fig. 10). At the urban site,
the total carbon budgets (Table 2) of FCP and FOP
were slightly different from those presented in
Fig. 6 due to the smaller amount of the data.
Table 2. The cumulative (non-gap filled) carbon budgets measured with closed-path (FCP) and open-path (FOP) analyzers corrected traditional way, with the MLR (FMLR), linear (FLin) and fitting methods (FFit) for urban and forest sites.
Differences between the closed-path and open-path fluxes expressed as percentages have also been listed. For
urban site, 5042 half-hour data between July 2007 and June 2008 while for forest site 1806 data points between
October and December 2006 were analyzed.
Urban
Forest
g C m–2
Percentage
g C m–2
Percentage
FCP
FOP
FMLR
FLin
FFit
1860
–
200
–
1530
–18
60
–67
1660
–11
90
–52
1830
–2
130
–30
1780
–4
180
–7
BOREAL ENV. RES. Vol. 14 •
Net CO2 flux comparison in an urban environment
At both sites, all heating corrections decreased
the difference between the cumulative CO2 fluxes
measured with closed-path and open-path analyzers. At the urban site, both the linear and fitting
methods corrected the NSE estimates well by
decreasing the difference between the analyzers to 30 g C m–2 (2%) and 80 g C m–2 (4%),
respectively (Fig. 10a). The MLR method failed
to correct the fluxes this well by decreasing the
difference to 200 g C m–2 (11%). For the forest
site, the best results were evidently obtained
with the fitting method, which decreased the
difference in NSE from 140 to 20 g C m–2, corresponding to a decrease from 67% to 7% (Fig.
10b). The linear and MLR methods decreased the
difference to 70 g C m–2 (30%) and 110 g C m–2
(50%), respectively. The difference between FCP
and FOP for the forest was large but comparable to
differences reported in other studies. Burba et al.
(2008) found a difference of 75%–81% between
the two analyzers outside the growing season.
Grelle and Burba (2007) showed a cumulative
plot for October in Sweden and got a difference
of 25% between the analyzers. We made the
cumulative plot for November only, presenting
similar weather conditions as in Sweden during
their measurement period, and found a difference
of 22% between FCP and FOP. Ono et al. (2008)
found that with near-zero closed-path flux, openpath showed 0 to –5.9 μmol m–2 s–1 above a
paddy field. Over long periods, this would cause
high difference between FCP and FOP.
Preferred surface heating correction
We calculated regressions between FCP and openpath CO2 fluxes corrected with different methods
511
for both urban and forest sites (see Table 3). All
heating correction methods improved the relationships for both sites, but the fitting method
gave the best results with R2 = 0.94, slope of 0.98
and offset of –0.25 at the urban site and R2 = 0.81,
slope of 0.97 and offset of –0.06 at the forest site.
Again, the linear method gave better results than
the MLR method by increasing the offset closer
to zero at both sites. The fitting method was the
only method removing most of the systematic
dependence of ΔFC on stability and wind direction at the urban site (not shown). Also the peak
seen in the vegetation sector was lowered, implying an underestimation of WPL partly due to the
sensor heating. However, the effect of vegetation
cover could not entirely be ruled out in the WPL
underestimation. None of the correction methods removed dependence of ΔFC on the wind
speed, being suggestive of other reasons for the
observed dependence as already discussed.
Besides the linear relationships and systematic dependencies, the fitting method corrected
most efficiently the fluxes on a diurnal scale for
both sites. Thus, we recommend that the fitting
method will be used in the surface heating correction despite the fact that the linear method
most efficiently corrected NSE for the urban
site. This can be a pure coincidence because
for the forest site the correction with the linear
trend was poor. The better functionality of the
fitting method as such is not surprising, since the
MLR and linear methods are based on equations
derived for a setup in which the open-path analyzer is mounted vertically. Sloping of the analyzer affects how much analyzers windows and
spars gain solar radiation, and how the thermal
boundary layer inside the optical volume develops. These have large effects on the surface tem-
Table 3. Linear regressions (p < 0.05) between closed-path and open-path CO2 fluxes corrected with different
methods for urban and beech forest sites in July 2007–June 2008 and October–December 2006, respectively. FOP
stands for traditionally corrected open-path CO2 flux, FMLR, FLin and FFit are the corrected open-path fluxes calculated
with the MLR, linear and fitting methods (see text for details).
FOP
Urban
Forest
Slope ± 95% CI
Offset ± 95% CI
R2
Slope ± 95% CI
Offset ± 95% CI
R2
0.98 ± 0.01
–1.40 ± 0.06
0.94
0.98 ± 0.02
–0.99 ± 0.08
0.80
FMLR
0.98 ± 0.01
–0.89 ± 0.06
0.95
0.96 ± 0.02
–0.67 ± 0.08
0.80
FLin
FFit
0.98 ± 0.01
–0.38 ± 0.06
0.94
0.93 ± 0.02
–0.35 ± 0.08
0.79
0.98 ± 0.01
–0.25 ± 0.06
0.94
0.97 ± 0.02
–0.06 ± 0.08
0.81
512
perature equations, and equations derived for the
vertical sensor do not work so well either. The
fitting method, on the other hand, does not have
any assumptions on the position of the analyzer
and provides a site-specific approach to correct
for the effect of sensor heating. The downside of
the fitting method is that it requires some period
of simultaneous measurements with open-and
closed-path analyzers. If there is no possibility to
use the fitting method, the linear method should
be used rather than the MLR method. Already
Burba et al. (2008) suggested the linear method
to be better in correcting long-term measurements.
Conclusions
The purpose of this study was to investigate
the differences between simultaneously-measured CO2 fluxes with the open- and closed-path
infrared gas analyzers (LI-7500 and LI-7000,
respectively) in an urban area of Helsinki over
one year from July 2007 to June 2008. The suitability of the analyzers to accurately measure
CO2 fluxes in such a complex environment and
the necessary correction procedures were studied. Besides, three different methods proposed
by Burba et al. (2006, 2008) to calculate the
effect of open-path surface heating were tested,
and comparisons were made with similar measurements made above beech forest in Denmark
between October and December 2006.
The co-spectra obtained from the Kaimal et
al. (1972) model were found to apply well to our
complex measurement site, even though originally the model was established for homogeneous
surfaces and flat terrain. The frequency response
corrections based on the model spectrum were
found to be 11% and 3% for closed-path and
open-path CO2 fluxes, respectively. The correlation between FCP and traditionally corrected FOP
was good with R2 = 0.94, slope of 0.96 and offset
of –0.61 μmol m–2 s–1, and showed the analyzers
to clearly measure the same signals. Despite the
apparent good correlation, the total carbon budgets for the studied year were 17% lower when
measured with the open-path analyser. Most of
the difference between the analyzers was caused
by warm (over 12 °C) daytime situations when
Järvi et al.
•
BOREAL ENV. RES. Vol. 14
the difference reached 25%, probably due to the
intense solar radiation. Beside the air temperature, the difference was systematically dependent
on the wind speed, wind direction and stability.
All tested surface heating corrections lowered the difference between the closed-path and
open-path CO2 fluxes at both urban and forest
sites. However, the best results were obtained
with the fitting method, for which a linear relationship between the surface and air temperature was derived with the bootstrapping method
from the measurements. The differences in total
carbon budgets measured with the closed-path
and open-path analyzers were reduced to 4%
and 7% at the urban and forest sites, respectively. The fitting method also removed most
efficiently the systematic dependencies of ΔFC.
The linear method gave the best results in carbon
budgets at the urban site by reducing the difference between FCP and FOP to 2%. However, for
forest it produced a difference of 30%, and failed
to correct for the diurnal pattern as efficient
as the fitting method for both sites. The MLR
method produced a difference of 11% and 50%
at the urban and forest sites, respectively. The
MLR and linear methods are based on surface
temperature equations which were derived for
measurement setups where the open-path analyzer was mounted vertically. Thus, they may
not work as well with sloping analyzers like in
our case. The fitting method, on the other hand,
provides a site-specific approach to correct the
effect of sensor heating, but the downside is the
need for simultaneous open-path and closed-path
measurements. If simultaneous measurements do
not exist (e.g. when correcting fluxes which date
back), the linear method should be preferred
over the MLR method.
Acknowledgements: For financial support we thank Maj and
Tor Nessling Foundation, the Academy of Finland Centre
of Excellence program (project nos. 211483, 211484 and
1118615), National technology Agency (TEKES, Ubicasting
project), Helsinki University Environmental Research Centre
(HERC), and EU projects IMECC and BRIDGE.
References
Aubinet M., Grelle A., Ibrom A., Rannik Ü., Moncrieff J.,
Foken T., Kowalski A.S., Martin P.H., Berbigier P.,
BOREAL ENV. RES. Vol. 14 •
Net CO2 flux comparison in an urban environment
Bernhofer Ch., Clement R., Elbers J., Granier A., Grünwald T., Morgenstern K., Pilegaard K., Rebmann C.,
Snijders W., Valentini R. & Vesala T. 2000. Estimates
of the annual net carbon and water exchange of forests:
The EUROFLUX methodology. Advances in Ecological
Research 30: 113–175.
Baldocchi D. 2003. Assessing the eddy covariance technique
for evaluating carbon dioxide exchange rates of ecosystems: past, present and future. Global Change Biology
9: 479–492.
Burba G.G, Anderson D.J., Xu L. & McDermitt D.K. 2006.
Correcting apparent off-season CO2 uptake due to surface heating of an open path gas analyzer: progress
report of the ongoing study. Proceedings of the 27th
Annual Conference of Agricultural and Forest Meteorology, San Diego, CA, American Meteorological Society,
Boston, MA, pp. 1–13.
Burba G.G., McDermitt D.K., Grelle A., Anderson D.J. & Xu
L. 2008. Addressing the influence of instrument surface
heat exchange on the measurements of CO2 flux from
open-path gas analyzers. Global Change Biology 14:
1–23.
Eugster W. & Senn W. 1995. A cospectral correction model
for measurement of turbulent NO2 flow. Boundary-Layer
Meteorol. 74: 321–340.
Eugster W., McFadden J.P. & Chapin F.S.III 1997. A comparative approach to regional variation in surface fluxes
using mobile eddy correlation towers. Boundary-Layer
Meteorol. 85: 293–307.
Foken T. & Wichura B. 1996. Tools for quality assessment of
surface-based flux measurements. Agric. For. Meteorol.
78: 83–105.
Grelle A. & Lindroth A. 1996. Eddy-correlation system for
long-term monitoring of fluxes of heat, water vapour and
CO2. Global Change Biology 2: 297–307.
Grelle A. & Burba G. 2007. Fine-wire thermometer to correct
CO2 fluxes by open-path analyzers for artificial density
fluctuations. Agric. For. Meteorol. 147: 48–57.
Grimmond C.S.B., King T.S., Cropley F.D., Nowak D.J.
& Souch C. 2002. Local-scale fluxes of carbon dioxide in urban environments: methodological challenges
and results from Chicago. Environmental Pollution 116:
243–254.
Ham J.M. & Heilman J.L. 2003. Experimental test of density
and energy-balance corrections on carbon dioxide flux as
measured using open-path eddy covariance. Agric. For.
Meteorol. 95: 1393–1403.
Horst T.W. 1997. A simple formula for attenuation of eddy
fluxes measured with first-order-response scalar sensors.
Boundary-Layer Meteorol. 82: 219–233.
Ibrom A., Dellwik E., Flyvbjerg H., Jensen N.O. & Pilegaard
K. 2007. Strong low-pass filtering effects on water
vapour flux measurements with closed-path eddy correlation systems. Agric. For. Meteorol. 147: 140–156.
Jones H.G. 1992. Plants and microclimate (2nd ed.). Cambridge University Press, Cambridge.
Järvi J., Hannuniemi H., Hussein T., Junninen H., Aalto P.
P., Hillamo R., Mäkelä T., Keronen P., Siivola E., Vesala
T. & Kulmala M. 2009. The urban measurement station
SMEAR III: Continuous monitoring of air pollution and
513
surface–atmosphere interactions in Helsinki, Finland.
Boreal Env. Res. 14 (suppl. A): 86–109.
Kaimal J.C., Wyngaard J.C., Izumi Y. & Coté O.R. 1972.
Spectral charateristics of surface-layer turbulence.
Quart. J. R. Met. Soc. 98: 563–589.
Kramm G., Dlugi R. & Lenschow D.H. 1995. A re-evaluation of the Webb correction using density-weighed averages. J. Hydrology 166: 283–292.
Leuning R. & King K.M. 1992. Comparison of eddy-covariance measurements of CO2 fluxes by open- and closedpath CO2 analyzers. Boundary-Layer Meteorol. 59:
297–311.
Liu H., Randerson J.T., Lindfors J., Massman W.J. & Foken
T. 2006. Consequences of incomplete surface energy
balance closure for CO2 fluxes from open-path CO2/
H2O infrared gas analyzers. Boundary-Layer Meteorol.
120: 65–85.
Massman W.J. 2000. A simple method for estimating frequency response corrections for eddy covariance systems. Agric. Forest Meteorol. 104: 185–198.
Massman W.J. & Lee X. 2002. Eddy covariance flux corrections and uncertainties in long-term studies of carbon
and energy exchange. Agric. Forest Meteorol. 113: 121–
144.
McMillen R.T. 1988. An eddy correlation technique with
extended applicability to non-simple terrain. BoundaryLayer Meteorol. 43: 231–245.
Moncrieff J.B., Massheder J.M., de Bruin H., Elbers J., Friborg T., Heusinkveld B., Kabat P., Scott S., Soegaard H.
& Verhoef A. 1997. A system to measure surface fluxes
of momentum, sensible heat, water vapour and carbon
dioxide. J. Hydrology 188–189: 589–611.
Moore C.J. 1986. Frequency response corrections for eddy
correlation systems. Boundary-Layer Meteorol. 37:
17–36.
Moriwaki R. & Kanda M. 2004. Seasonal and diurnal fluxes
of radiation, heat, water vapour, and carbon dioxide over
a suburban area. J. Appl. Meteorol. 43: 1700–1710.
Nemitz E., Hargreaves K.J., McDonald A.G., Dorsey J. &
Fowler D. 2002. Micrometeorological measurements of
the urban heat budget and CO2 emissions on a city scale.
Environ. Sci. Technol. 36: 3139–3146.
Nobel P.S. 1983. Biophysical plant physiology and ecology.
Miller Freeman and company, New York.
Ono K., Miyata A. & Yamada T. 2008. Apparent downward CO2 flux observed with open-path eddy covariance
over a non-vegetated surface. Theor. Appl. Climatol. 92:
195–208.
Pilegaard K., Mikkelsen T.N, Beier C., Jensen N.O., Ambus
P. & Ro-Poulsen H. 2003. Field measurements of atmosphere-biosphere interactions in a Danish beech forest.
Boreal Env. Res. 8: 315–333.
Rannik Ü. & Vesala T. 1999. Autoregressive filtering versus
linear detrending in estimation of fluxes by the eddy
covariance method. Boundary-Layer Meteorol. 91: 259–
280.
Rannik Ü., Keronen P., Hari P. & Vesala T. 2004. Estimation
of forest-atmosphere CO2 exchange by eddy covariance
and profile techniques. Agric. Forest Meteorol. 126:
141–155.
514
Rannik Ü., Kolari P., Vesala T. & Hari P. 2006. Uncertainties in
measurements and modelling of net ecosystem excahnge
of a forest. Agric. Forest Meteorol. 138: 244–257.
Rogiers N., Conen F., Furger M., Stöckli R. & Eugster
W. 2008. Impact of past and present land-management on the C-balance of a grassland in the Swiss
Alps. Global Change Biology, doi: 10.1111/j.13652486.2008.01680.x.
Roth M. 2000. Review of atmospheric turbulence over cities.
Q. J. R. Meteorol. Soc. 126: 941–990.
Roth M. & Oke T.R. 1993. Turbulent transfer relationships
over an urban surface. I: Spectral characteristics. Q. J. R.
Meteorol. Soc. 119: 1071–1104.
Salmond J.A., Oke T.R. & Grimmond C.S.B. 2005. Venting
of heat and carbon dioxide from urban canyons at night.
J. Appl. Meteorol. 44: 1180–1194.
Serrano-Ortiz P., Kowalski A.S., Domingo F., Ruiz B. &
Alados-Arboledas L. 2008. Consequences of uncertainties in CO2 density for estimating net ecosystem CO2
exchange by open-path eddy covariance. BoundaryLayer Meteorol. 126: 209–218.
Järvi et al.
•
BOREAL ENV. RES. Vol. 14
Schotanus P., Nieuwstadt, F.T.M. & De Bruin H.A.R. 1983.
Temperature measurement with a sonic anemometer and
its application to heat and moisture fluxes. BoundaryLayer Meteorol. 26: 81–93.
Suyker A.E. & Verma S.B. 1993. Eddy correlation measurement of CO2 flux using a closed-path sensor: theory and
field tests against an open-path sensor. Boundary-Layer
Meteorol. 64: 391–407.
Vesala T., Järvi L., Launiainen S., Sogachev A., Rannik Ü.,
Mammarella I., Siivola E., Keronen P., Rinne J., Riikonen A. & Nikinmaa E. 2008. Surface-atmosphere interactions over complex urban terrain in Helsinki, Finland.
Tellus 60B: 188–199.
Vogt R., Christen A., Rotach M., Roth M. & Satyanarayana
A. 2006. Temporal dynamics of CO2 fluxes and profiles
over a central European city. Theor. Appl. Climatol. 84:
117–126.
Webb E.K., Pearman G.I. & Leuning R. 1980. Correction
of flux measurements for density effects due to heat
and water vapour transfer. Q. J. R. Meteorol. Soc. 106:
85–100.